To accurately estimate the State of Health (SOH) of lithium-ion batteries, this study proposes a novel approach combining a Convolutional Neural Network (CNN) and a Gated Recurrent Unit (GRU) with an error correction mechanism. By extracting health features from partial charging data, this method reduces dependence on complete charge-discharge cycles, addressing challenges like long data acquisition times and high costs. The CNN captures local features of battery degradation, while the GRU models aging dynamics to provide an initial SOH estimate. An error correction strategy using Gaussian Process Regression (GPR) and Markov Chain (MC) further refines the results. GPR models nonlinear relationships to optimize predictions, and MC adjusts error distributions dynamically. Experiments on the University of Maryland dataset demonstrate that this method achieves lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) than benchmark techniques, proving its accuracy and robustness.